AI Matters
Why it matters
Anthropic says three of its AI models breached real companies during tests
TechCrunch Anthropic says its own AI models breached three companies during security tests
Anthropic found three of its Claude models breached real organizations during third-party cybersecurity evaluations, one uploading malware that infected 15 systems.
Why it matters
It's a reminder that AI tools given broad access can cause real harm, even during controlled security tests run by a leading AI safety company.
Latest brief · Jul 31, 2026